From Quarter to All: Accelerating Speculative LLM Decoding via Floating-Point Exponent Remapping and Parameter Sharing

Documenti analoghi

• MCBP: A Memory-Compute Efficient LLM Inference Accelerator Leveraging Bit-Slice-enabled Sparsity and Repetitiveness
  di: Wang, Huizheng, et al.
  Pubblicazione: (2025)
• SOFA: A Compute-Memory Optimized Sparsity Accelerator via Cross-Stage Coordinated Tiling
  di: Wang, Huizheng, et al.
  Pubblicazione: (2024)
• BBAL: A Bidirectional Block Floating Point-Based Quantisation Accelerator for Large Language Models
  di: Han, Xiaomeng, et al.
  Pubblicazione: (2025)
• SpecMamba: Accelerating Mamba Inference on FPGA with Speculative Decoding
  di: Zhong, Linfeng, et al.
  Pubblicazione: (2025)
• Designing Spatial Architectures for Sparse Attention: STAR Accelerator via Cross-Stage Tiling
  di: Wang, Huizheng, et al.
  Pubblicazione: (2025)